Robust social categorization emerges from learning the identities of very few faces.

Abstract

Viewers are highly accurate at recognizing sex and race from faces -though it remains unclear how this is achieved. Recognition of familiar faces is also highly accurate across a very large range of viewing conditions, despite the difficulty of the problem. Here we show that computation of sex and race can emerge incidentally from a system designed to compute identity.We emphasise the role of multiple encounters with a small number of people, which we take to underlie human face learning. We use highly variable everyday 'ambient' images of a few people to train a Linear Discriminant Analysis (LDA) model on identity. The resulting model has human-like properties, including a facility to cohere previously unseen ambient images of familiar (trained) people -an ability which breaks down for the faces of unknown (untrained) people. The first dimension created by the identity-trained LDA classifies both familiar and unfamiliar faces by sex, and the second dimension classifies faces by race -even though neither of these categories was explicitly coded at learning. By varying the numbers and types of face identities on which a further series of LDA models were trained, we show that this incidental learning of sex and race reflects covariation between these social categories and face identity, and that a remarkably small number of identities need be learnt before such incidental dimensions emerge. The task of learning to recognise familiar faces is sufficient to create certain salient social categories.

Face averages and multiple images in a live matching task

“…We created the average images by first deriving the shape of each image using a semi‐automatic landmarking system designed to register 82 points on the face aligned to anatomical features, where only five locations are selected manually (for details, see Kramer, Young, Day, & Burton, ). Each average was created by warping the 12 images of the model to the average shape of those 12 images, and then calculating the mean RGB colour values for each pixel using the InterFace software package (Kramer, Jenkins, & Burton, ).…”

Enhancing CCTV: Averages improve face identification from poor‐quality images

“…Each average was created by warping the 10 images of an identity to the average shape of those 10 images and then calculating the mean RGB colour values for each pixel. The unpixelated images were landmarked using our semi‐automatic system (where only five locations are selected manually—for details, see Kramer, Young, Day, & Burton, ). After pixelation, the images were again landmarked using the system.…”

Interaction between social categories in the composite face paradigm.

“…Although we were able to vary gender and occupation orthogonally in the sense that male or female faces could have different occupations, these characteristics are none the less linked through the face's identity. In fact interactions between gender and identity are evident in data from face learning tasks (Baudouin & Tiberghien, 2002) and statistical analyses of face images show that gender can be derived incidentally from learning to categorise face identity (Kramer, Young, Day & Burton, 2017).…”